Comprehensive correlation analysis enabled neural network prediction of heat and mass transfer during gas hydrate decomposition

Abstract

A significant amount of natural gas is stored in a form of hydrate. Yet commercial exploitation of natural gas hydrate remains quite challenging due to limited comprehension of internal heat and mass transfer processes. In this work, a numerical model is developed to describe heat and mass transfer during methane hydrate decomposition and to provide sufficient data for neural network modeling. Based on the numerical model, the temporal and spatial evolution patterns of several decomposition characteristics, including multiphase saturation, temperature, gas pressure, and gas velocity, are elucidated. More importantly, the effects of 19 types of variables related to various boundary conditions, physical properties, and initial conditions are comprehensively investigated. A comprehensive correlation map between these variables and four key heat and mass transfer parameters reveals 41 positive and 35 negative correlations. Driven by abundant simulation data, an artificial neural network model is then developed to predict the heat and mass transfer parameters. As validated, the neural network model shows satisfactory efficiency and accuracy, achieving relative errors below 2% in the prediction of various heat and mass transfer parameters. This study provides a comprehensive theoretical guide and a useful method for understanding, regulating, and optimizing the natural gas hydrate exploitation.